Lecture II-18: N = 2 Super-symmetric Yang-Mills theories

in dimension four: Part II

Edward Witten

I Review of material from the last lecture

We are considering pure N = 2 super-symmetric gauge theory (i.e., without hyper-multiplets) in
four-dimensional Minkowski space. The gauge group is either SU(2) or the closely related group
SO(3). (In fact, working as we are today over Minkowski four-space, which is contractible, we
will not see the distinction between these two gauge groups.) We write the theory in the N = 1
super-symmetric notation. The elds which form a single N = 2 vector multiplet decompose
under the N = 1 super Poincare group in four dimensions as a vector multiplet A (the gauge
elds) and a chiral super-eld
 = +  +



where  is a scalar eld (under the Poincare group). It is a section of the complexication of the
adjoint bundle. The microscopic Lagrangian is
Z Z
L = d4xd20 (0)TrW W  + c:c: + d4 xd2Im0 (0)Tr()
where W is a certain piece of the super-connection of the super-connection A. (Here 0 is a
cut-o used to render nite the Feynmann path integrals. We vary 0 as a function of 0 in
order to renormalize, i.e., render the expectation values in the theory nite as 0 7! 1. In order
to decide how to renormalize we must x a so-called `subtraction point'  at which to x the
value of one parameter in the theory. We shall see later a convenient normalization scheme for
this particular theory which involves the details of the theory (the u-plane, etc). But for now
 Notes by John Morgan

1
we leave vague the actual renormalization scheme. Once we have xed the value of the extra
parameter , the coupling constant 0 is a function of 0 .) While  depends on the cut-o, it is
independent of the elds.
Our object today is to solve this theory by explicitly understanding the moduli space M of
quantum vacua, as well as a description of the theory near each vacuum state. As we established
last time, the space of classical vacua of this theory consists of gauge equivalence classes of
constant matrices in su(2) !
1 a 0
p
2 0 a
modulo the action of the Weyl group, which is a group of order two switching a and a. This
space is analytically isomorphic to C and the natural local holomorphic coordinate to use on
this plane is u = Tr2 = a2 . While the analytic space of classical vacua is smooth, there is one
point in the moduli space of classical vacua where the Kahler metric becomes singular. This
is the origin. The singularity is a consequence of the fact that at this point there are more
massless elds (in fact a super-symmetric SU(2)-theory is the lower energy eective theory).
Furthermore, as we discussed last time, away from u = 0 the low energy eective theory is a
pure N = 1 super-symmetric U(1)-gauge theory. Thus, away from u = 0, the nature of this
eective theory is described by a family of tori over the punctured u-plane. In the case of the
moduli space of classical vacua this is a family of tori with constant j invariant given by the
choice of  in the Lagrangian. The monodromy around innity for this family is 1. The Kahler
metric on the u-plane is
ds2 = Im dudu
4juj2 :
(Notice that there is a singularity in the metric at u = 0 as predicted by the fact that at this
vacuum there are extra massless elds classically.)

I.1 First results about the moduli space M of quantum vacua

Let M be the space of quantum vacua. Notice that, because we are assuming that our theory
is N = 2 super-symmetric, the space of quantum vacua has a Kahler metric (from the weaker
N = 1 super-symmetry) and over this space of vacua there is a holomorphic family of tori. The
eld u, which is a coordinate on the space of classical vacua, is a chiral eld in the theory, and
thus its expectation values in the various quantum vacua denes a holomorphic function on M.

2
We denote this function by u: M ! C:
u(
) = h
juj
i; for all
2 M:
The part of M that is most closely related to the u-plane of classical vacua is the region
where u is large. More precisely, suppose that we have xed the mass parameter, , in the
renormalization scheme which determines the specic quantum theory we are studying. Suppose
that we are studying the theory near a classical vacuum
with u(
) >> 2. (Notice that since
u has dimensions of mass squared it naturally compares to 2.) Because our SU(2)-theory is
asymptotically free, and the U(1)-theory to which it is limiting is free in the infra-red, if u >> 2,
then the quantum theory is well-approximated near
by the classical theory. This means that
the low energy quantum theory near
is described by a sigma model on the u-plane, perhaps
perturbed by some super-potential. But N = 2 super-symmetry means that any such super-
potential must be a constant, and hence cannot lift the vacuum degeneracy near
. (There is also
the possibility of a Fayet-Iliopoulos -term to consider as well, but the SU(2)R -symmetry rules
these out.) What this argument shows is that for large u there is a unique quantum vacuum for
each classical vacuum. That is to say, there is a constant C() > 0 and an open subset U  M
with the property that u: U ! (fz 2 Cjjz j > C g) is a holomorphic isomorphism.
In Seiberg's lectures we have seen other examples of gauge theories where near innity in the
space of classical vacua the gauge group SU(Nc ) is broken down to a non-abelian gauge group
SU(Nc Nf ). Since the gauge theory for this smaller group is not free in the infra-red, we cannot
conclude that the space of quantum vacua near innity can be identied with the subspace of
critical points of a superpotential on the space of classical vacua. In fact, in these examples the
moduli space of quantum vacua is a nite covering space of the space of classical vacua near
innity. To repeat, this is possible because the low energy theory is not free in the infra-red.
When it is free, as in the case under discussion now, and when N = 2 super-symmetry rules out
the existence of super-potentials, the classical and quantum moduli spaces can be identied over
a neighborhood of innity in the space of classical vacua.
One assumption that we shall make without any justication is that the moduli space M is
complete in an appropriate sense so that it corresponds to an open complex algebraic variety.
We have found one component of M which has an end isomorphic under u to a neighborhood
of innity in C. Two other questions present themselves:

3
  Are there other `ends' of M, ends that are not seen classically?
 Are there other irreducible components of M not seen classically?
The answer to the rst question is `No.' The point is that all ends of the quantum moduli
space must be visible classically, i.e, they must be related to ends of the classical moduli space.
The reason is that quantum eects are bounded and hence cannot eect a non-proper change in
the passage from the space of classical vacua to the space of quantum vacua.
As to the second question, there are no examples known where there are branches of the
quantum space of vacua which are not seen classically. Still in this generality, it has not been
established that such components do not exist. Nevertheless, in our case we can give arguments
showing that there are no other components. The rst thing to note is that, by the answer to
the rst question, any other component of M must be compact. To exclude the possibility of
other compact branches we use the fact that the theory has a relevant N = 1 super-symmetric
perturbation. That is the perturbation
Z

L = d4xd2u + c:c:

where  is a complex constant. As we saw in Lecture II-??, with this perturbation the resulting
SU(2)-theory has two vacua, each with a mass gap and each with connement. If we examine
what happens to these perturbed theories in the low energy limit, then of course they look the
same: they have two vacua each with a mass gap and connement. Thus, this perturbation
would have to lift the vacuum degeneracy along a compact component N of M by adding a
super-potential. On the other hand, since u is a holomorphic function along this branch, it is
constant. Now the super-potential term that we have added to the microscopic Lagrangian is
simply u which is constant along this branch. One can see that this persists in the low energy
theory, so that the low energy eective super potential is also a multiple of u, and hence is
constant. This means that the addition of the above super-potential cannot lift the vacuum
degeneracy along this branch. Since we know the resulting low energy theory has only two
vacua, this implies that the other branch is in fact not present. (This argument still allows for
the possibility of isolated points (at most two), but we shall see later by counting singularities,
that they do not occur either.)
We have now argued that M has a single branch (plus possibly at most two isolated points)

4
and that this branch has a single neighborhood of innity which is mapped by u isomorphically
to a neighborhood of innity in the u-plane. By our completeness assumption, we know that
M is a possibly singular, irreducible complex curve with one puncture. The existence of the
holomorphic function u with a only one pole, that being the simple pole at innity of M, implies
that M is actually the complex plane and that u is a global holomorphic parameter on M.
Following the analysis of the last lecture, our goal is to solve the theory by nding the family of
elliptic curves (in Weierstrass form) over M minus the singular points

y2 = x3 + A(u)x + B(u)

and a holomorphic two-form  on the total space of this family

 = f(u)du dx
y
whose periods determine the dierentials da and daD of the dual pair of local holomorphic
coordinates.

I.2 The nature of innity in M

As we have already remarked, we obtain information near innity in M, i.e., where u is large
using asymptotic freedom of the SU(2)-theory. Our goal here is to compute the monodromy at
innity of the family of elliptic curves over the moduli space of quantum vacua. We begin by
studying the monodromy at innity for the space of classical vacua. Recall that the space of
classical vacua is the space of matrices
!
1 a 0
= p
2 0 a
up to a 7! a. The dual coordinate aD is given by aD = cl: a where cl: is a constant (denoted by
0 earlier) with large imaginary part. The global coordinate on the u-plane is u = Tr(2) = a2 ,
so that a = pu. From this it is clear that the monodromy at innity in this classical family is
1.
The monodromy around innity for the space M of quantum vacua is dierent. We give two
arguments calculating it. The rst is based on asymptotic freedom and the second is based on
the U(1)R -anomaly. To compute e (u) for large u we consider the 1-loop formula. With an

5
appropriate choice of the mass parameter , the formula is:
 
Im = g42 = 4 2b0 ln( juj=) ;
p
(1)
e
where b0 is the 1-loop -function coecient.
One loop computations identify
b0 = 41 2 :
Since  is holomorphic, knowing its imaginary part determines it uniquely. We see that for u
large we have
2)
e = 2iln(u=
2 : (2)
Thus, we see that monodromy around innity shifts  by 2 and hence shifts  by 4. This
implies that the monodromy around innity in M is given by
!
1 2
:
0 1
Said another way we are considering the system
u2 = a
daD =  = 2iln(u=2) (3)
da 2
as we turn once around innity. Clearly, a comes back to a and  comes back to  2. The
above form for the monodromy then follows easily.
Now let us compute this monodromy using the anomalous U(1)R symmetry. Classically, the
U(1)R -charges of the component elds in the gauge super-multiplet are given as follows: the
charge of A is zero, the charges of ; are 1 and the charge of a is two. This means that u = a2
has charge 4. (This is related to the fact that in Donaldson theory (pt) is four-dimensional.)
This means that under the U(1)R symmetry ei the variable u transforms by u 7! ue4i . In
fact, the sign is negative. On the other hand, as we know, there is a quantum anomaly breaking
this U(1)R down to Z=8Z (which is the dimension of the one-instanton moduli space on R4 ).
This is an index computation which uses the fact that the Casimir of the adjoint representation
of su(2) is 4 times the Casimir of the dening representation. This means that under the U(1)R
symmetry ei the angle  transforms to  8. Thus, as we transverse a loop around innity
in the u-plane the argument of u changes by 2,  increases by 2=4 = =2 and  decreases by
4. This means that  7!  2 under the monodromy around innity.

6
 Of course, these are not independent computations. Super-symmetry imposes a relation
between the U(1)R anomalies and the -function.
Actually, we can say more about the nature of  at innity. For juj >> 2 in perturbation
theory we have a series
2)
(u) = 2iln(u=
2 + 
2 (4)
where the rst term is the sum of the classical value and the one-loop correction (as calculated
above) and the second term is the sum of the corrections for higher loops. These higher loop
corrections produce a power series in g2 (k-loop corrections produce the power g2k 2). Since there
are no loop corrections to  and since  is a holomorphic function of u, it follows immediately
that all the higher loop corrections vanish. Thus, in perturbation theory the one-loop correction
is exact. There are however non-zero corrections to the perturbation expansion coming from
instantons. Of course, these can not be seen in perturbation theory. Let's study their nature. We
know that  is invariant under the U(1)R -symmetry. An instanton breaks the U(1)R -symmetry
by 8k where 82k is its total energy. Since u has U(1)R -charge 4, the k instantons go with
even u2k. Of course, the dimensionless expression is 4k =u2k. Thus, the instanton action for an
instanton of total energy 82 k is of the form
4k
fk exp( 8k2=g2 )eki = fk 4ke2ki = fk u2k
for a dimensionless constant fk . Thus, the full expression for  at juj >> 1 is of the form
2 ) X  4k
(u) = 2iln(u=
2 + k1 fk u2k : (5)

I.3 BPS states and singularities in M

Next, we wish to understand the singularities in M. By a singularity we mean a point of the
moduli space where the physics changes. This may or may not be accounted for by a topological
singularity in the space. As a general principle, we expect that the singularities have to do
with particles becoming massless. In this case since the particles are to add singularities to the
low energy limit which is a U(1)-theory, they must be charged under the U(1). It is also quite
reasonable to expect that if any particles become massless, then some BPS states (which are
the states of lowest energy) become massless. In our case we have two types of BPS states near
innity in the u-plane. We have the vector multiplet (i.e., the gauge boson) which has ne = 1

7
and nm = 0 and we have the magnetic monopoles with nm = 1 and any value of ne . (Of course,
as we increase  by 2 a particle with nm 6= 0 has ne which changes by 2, so that having one
BPS state with nm = 1 gives rise to BPS states with all even or all odd values of ne. Our claim
here is that there are two families of these BPS states, one with even ne and one with odd ne .)
Notice that this BPS spectrum is invariant under the monodromy at innity since under this
monodromy ne 7! ne + 2nm and nm 7! nm . On the other hand we claim that this spectrum
can not be invariant under the entire monodromy of our family. The reason is that as we shall
see the image of the monodromy on all of M is a non-abelian subgroup of SL2 , under which the
above-described BPS spectrum is not invariant. This means that the BPS spectrum cannot be
continuous throughout M, that is to say there must be a jumping locus, which is a real curve in
M along which extra BPS states become massless. As we saw in the last lecture, this happens
when a=aD becomes real.
As we have seen M is a complex analytic space with exactly one irreducible component.
Furthermore, the holomorphic map u: M ! C is proper and is an isomorphism in a neighborhood
of innity of M. Next, we wish to analyze the singularities of M. We are assuming that M is
in fact a complex algebraic variety. There must be singularities in M since the monodromy at
innity is non-trivial. As we have argued singularities in M are associated with a certain super-
symmetric multiplet consisting of charged BPS states becoming massless. We have two types
of BPS multiplets { those where the maximal spin in the multiplet is 1 (the vector multiplet)
and those where the maximal spin is 1=2, (the hyper-multiplet). A vector multiplet becoming
massless corresponds to restoring a non-abelian gauge group (SU(2)). There are several reasons
one does not expect this to happen. First of all, with unbroken SU(2) one has a theory which is
unstable in the infra-red. This is exactly what happens at the singularity in the classical moduli
space. Since the -function is negative, we have owed away this theory, so we don't expect to
see it occurring in the low energy limit which we are assuming to be free in the infra-red. Of
course, if our gauge theory has enough hyper-multiplets becoming massless to make  > 0 then
we might see such a low energy limit. But at these points we have multiplets with maximal spins
both 1 and 1=2 becoming massless. So the simplest thing that we can do is to require that only
a hyper-multiplet with maximal spin 1=2 become massless. Let us analyze this possibility.
Suppose that the singularity in M is associated to a multiplet whose maximal spin is 1=2
becoming massless. Recall that a massive N = 2 super-symmetric multiplet of spin at most 1=2

8
is automatically BPS saturated. Thus, we might expect that what is happening is that one of
the BPS states that exist in theory for juj >> 2 is simply becoming massless at the singularity.
Of course, more complicated phenomena could occur. For example, as we come in from innity
in M we could cross a jumping line of the BPS spectrum before we reach the singularity, and
then at the singularity it is one of the `new' BPS saturated states that becomes massless, rather
than one of the original states that exist near u = 1.
Let us assume that a hyper-multiplet becomes massless. Denote its charge by (ne ; nm) under
the electric and magnetic charge. We have already argued that this hyper-multiplet must be
charged, so that q = g:c:d (ne ; nm ) is a well-dened positive number. After an SL2 (Z) change
in the representation of the theory by a Lagrangian, we can arrange that ne = q and nm = 0.
Thus, near the singular point u0 2 M our theory is a U(1)-theory with a charged particle which
is becoming massless at u = u0. (But notice that it may well be the case that nm 6= 0. If that
happens then the U(1)-gauge elds in the representation of the theory near u = u0 are dierent
from the U(1)-gauge elds that we naturally have at innity.) Unlike the spin-one case, this
theory is stable in the infra-red, i.e.,  > 0. Since the behavior of the theory near u = u0 is
infra-red free, we can compute ge and e in this region using perturbation theory. We see
that ge vanishes logarithmically and is given by the one-loop -function. The relevant one-loop
Feynmann diagram is:

q q

Clearly, this diagram gives a multiplicative factor of q2 . It also diverges logarithmically for
u 7! u0. The result is that for u near u0 we have:
(u) = 2q2 ln(u u02i) + constant :

This means that the monodromy of (u) in PSL2 (Z) around this singularity is
!
1 2q2
 : (6)
0 1
9
 Of course, the monodromy is actually an element in SL2 (Z) and is only determined up to
sign by the monodromy in PSL2 (Z) To pin down the sign we argue as follows. The physics
near u = u0 is described by a U(1)-gauge theory with gauge elds (A; ~ ~a) and a charged hyper-
multiplet (T; T 0 ) which is becoming massless at u0. Thus, the Lagrangian is of the form of the
gauge couplings plus a super-potential a~qTT 0 . The scalar eld ~a is a local coordinate on the u-
plane and in fact is related to a; aD at innity by an ISL2 (Z) transformation. (The image of this
transformation in SL2 (Z) is determined by the charge (ne ; nm ) of the hyper-multiplet (T; T 0 )
with respect to the U(1) at innity in the u-plane.) Since the U(1)-theory exists throughout
this neighborhood even at the singularity, we see that under monodromy around u0 the eld a~
comes back to itself. This implies that there is an invariant vector under the SL2 (Z)-monodromy
around u0 and hence shows that the monodromy is as written in Equation 6 with a plus sign.
We can generalize this picture. Suppose that at u0 there are several hyper-multiplets of
charge qi becoming massless. Then the local form of  would be
X 2 ln(u u0 ) + constant
 =2 qi 2i
i
and the monodromy around u0 would be
P !
1 i 2qi2
:
0 1
Notice that the monodromy around a singularity where a hyper-multiplet is becoming mass-
less has trace 2 whereas the monodromy at innity has trace 2. This means in particular, that
there cannot be only one singularity in the u-plane if the singularity is of this type. (This of
course also follows from the positivity of  in the complement of the singularities.)

I.4 The number of singularities in M

To understand the number of singularities in M, let us begin with a microscopic N = 2 super-
symmetric SU(2)-theory and break it to an N = 1 super-symmetric theory by adding a pertur-
bation of the form Z

L = d4xd2mu + c:c; (7)
with m a complex constant. (Recall that u = Tr2 .) This theory has two vacua, each with a
mass gap and with connement.

10
 Let us suppose that at a point in the space of quantum vacua we have Nh hyper-multiplets
becoming massless, say (Ti ; Ti0) of charge qi . So near this point the physics is described by a
super potential !
X 0
W = a~ qiTi Ti + mu(~a):
i
where, as before a~ is a local parameter on the u-plane and is related by an SL2 (Z) transformation
to a; aD at innity. Let us nd the space of vacua in the macroscopic theory. That is to say we
solve dW = 0 and divide by the C -action. Clearly, the equation dW = 0 is equivalent to

~aTi = a~Ti0 = 0 for all i:

X
m du
d~a + qi Ti Ti0 = 0:
I
If Nh > 0 this gives us a positive dimensional space of vacua. This is impossible since the
microscopic theory has only two vacua. This means that (as indicated above) at each singularity
we have only one hyper-multiplet becoming massless. Furthermore, since du da~ 6= 0 we see from
the second equation that T; T 0 6= 0. By the rst equation, this means that ~a = 0. Since there is
only one hyper-multiplet, the equation becomes

m du 0
d~a ja~=0 + qTT = 0:
This determines the vacuum uniquely up to the action of C . Thus, we see that there is a
unique vacuum associated to this singularity. Having one vacuum for each singularity in the
macroscopic theory, and having only two vacua in the microscopic theory, it follows that there
are at most two singularities in M. Notice also that the critical points for u on M also give
vacua in the macroscopic theory. Thus, we see

Nc + Nh  2

where Nc is the number of critical points in M of the analytic function u and Nh is the number
of points at which a hyper-multiplet becomes massless.
As we have already remarked, the monodromy around a point where a hyper-multiplet is
massless and the monodromy at innity have opposite traces and hence are not conjugate in
SL2 (Z). At a critical point of u there is no monodromy. Thus, these facts about monodromy
imply that Nh  2. It follows that Nh = 2 and Nc = 0. That is to say u: M ! C is

11
everywhere a local analytic isomorphism. Since it is an isomorphism between neighborhoods
of innity, it follows that u: M ! C is an isomorphism. (Recall that we established this fact
directly from the fact that u is an isomorphism at innity. This gives a consistency check on
our model.) Even though the spaces of classical and quantum vacua are identied, there are
important dierences. Whereas classically, there is only one singular point in the u-plane and at
this singular point the SU(2)-theory is restored as the low energy limit, quantum mechanically
there are two singularities in the u-plane (at as yet to be identied points), and at each of these
a charged spin 1=2 hyper-multiplet becomes massless. Furthermore, unlike the classical case,
the monodromy in the quantum family is non-trivial in PSL2 (Z). This implies that, unlike
the classical case, the family of tori over the space of quantum vacua do not have constant
j-invariant.
Notice that now that we have found two singularities in the u-plane, each contributing a
vacuum when we add the perturbation in Equation 7. Since there are at most two vacua in the
perturbed theory, it follows that there cannot be isolated points of the space of quantum vacua;
these isolated points would also produce quantum vacua of the perturbed theory.
The U(1)R -symmetry is broken down to a Z=8Z-symmetry which since u has charge four
under U(1)R acts on the u-plane as a Z=2Z-symmetry. It is the symmetry u 7! u. This implies
that the two singularities in M are at points u0 for some u0 6= 0. As we have already seen
dimensional analysis tells us that u goes like 2 , the square of the mass parameter in the theory.
This means that we can make a choice of the mass parameter  so that the quantum theory
labeled by  has singularities at u = 2 . In this way we are adopting the precise denition
of the mass scale parameter  promised in the introduction. Notice that this choice of  is
dierent from the one that we have used until now. The only aect of this change in the choice
of  simply is to introduce an additive constant into Formulas 1 through 5.

I.5 The new massless particles

The next step is to identify what is becoming massless. Let (ne ; nm ) be the charge of the
hyper-multiplet becoming massless at 2 when we express things in the basis in which the
monodromy at innity is upper triangular. If nm = 0, then the monodromy at 2 commutes
with the monodromy at innity. Since the total space of the non-singular part of the family is the

12
three-times punctured sphere, the monodromies at 2 and 1 generate the entire monodromy
image. Thus, if the monodromy at 1 and 2 are simultaneously upper triangular, then the
entire monodromy image is upper triangular and hence commutative. This would mean that
Im could not be positive everywhere. Our conclusion is that nm 6= 0. That is to say, the
particle becoming massless at u = 2 is magnetically charged. Exactly the same analysis holds
for the singularity at 2. This is a very interesting phenomenon. Particles which started life
when juj >> 2 as massive solitons are becoming massless as u approaches 2. Also, it means
that our description of the theory near 2 as a U(1)-theory is a dierent representation from
the U(1)-representation of the theory near innity in the u-plane.
Now we wish to calculate the magnetic charge nm of these particles. Notice that as we come
in from innity nm is well-dened independent of the path we choose from innity to 2 but
the electric charge depends on the path { if we wind around the pair of singularities ne changes
by twice the magnetic charge, since the monodromy around innity is ne 7! ne + 2nm .
The model we are discussing today has various mathematical applications, but just in terms
of physics one of its most striking applications was to give a new model of connement. We
recall that connement is the statement that there is a linear potential between external electric
charges or alternatively that there is area law decay for a Wilson loop in the an appropriate
representation (in the present example, the two-dimensional representation of SU(2)). We recall
that in a Higgs phase, where a charged eld gets a vacuum expectation value1, one can explicitly
calculate by a topological argument that the 't Hooft loop gets area law decay. If one could just
naively exchange electricity and magnetism, then Wilson loops would be exchanged with 't Hooft
loops and we would conclude that connement will arise if a magnetically charged eld gets a
vacuum expectation value. This was argued heuristically in the 1970's by 't Hooft, Mandelstam,
Nambu and others but with great diculty in exhibiting actual models, since usually one does
not have magnetically charged elds in the formalism and magnetically charged objects (which
arise as classical solutions) are heavy in the weakly coupled regime where one understands them
concretely. Thus it was generally rather hard to see how one could actually exhibit a phenomenon
in which a magnetically charged eld gets a vacuum expectation value.
In the present context, this happens. We have already seen that after adding the mu per-
turbation to the superpotential, which is expected to give connement, the vacuum near u = 2
1 This is a somewhat imprecise description as we have seen in previous lectures, but is valid in weak coupling.

13
has T; T 0 6= 0, so in fact in this vacuum (nm being non-zero) a magnetically charged object has
obtained a vacuum expectation value. Thus as expected a magnetically charged eld obtained a
vacuum expectation value just in the conning case. (To pursue this with greater precision, it is
important that the magnetically charged eld in question has nm odd. Indeed, as the center of
SU(2) is Z=2Z, the explanation of connement via a dual of the usual Higgs mechanism requires
a nonzero expectation value of a eld of odd nm . We will see that nm actually is odd, in fact
1, for T and T 0 .)
Notice something surprising has happened. The magnetic monopoles have mass at innity
p p
on the order of jujln(juj) whereas there are other BPS states of masses on the order of juj.
Thus, at innity the monopoles are not the least massive of the massive particles. Nevertheless,
they are the ones becoming massless at the singular points.

I.6 Explicit nature of the family of elliptic curves

We are now ready to describe the macroscopic theory associated to the quantum theory labeled
by mass parameter  by describing explicitly the family of elliptic curves and the two-form 
associated with the space of quantum vacua of this theory. As we have seen, for this theory
the moduli space of quantum vacua is C with parameter u = Tr2. The SL2 (Z)-monodromy at
innity for the family is !
1 2
:
0 1
There are in addition two other points 2 in the u-plane where there are singularities. The
SL2 (Z)-monodromy, m2 at each of these points has trace two. We write
!
1 1 + b b2 =a
m2 = ;
a 1 b
for integers a; b. (This is the general form of a matrix of trace two and determinant one.) Since

m 2 = m21 m1

has trace two, we conclude that

! !
1 + b b2=a 1 2
2 = Tr :
a 1 b 0 1

14
 It follows by a direct computation that a = 2. Since b2=a is an integer, we see that b is
even. Now let us conjugate the matrix for m21 by the matrix
!
1 1
;
0 1
which commutes with the monodromy at innity. The result is to change b by 2. Thus, after a
number of these conjugations we arrive at b = 0. This means that in an appropriate basis the
monodromy at innity is !
1 2
0 1
2
and the monodromy at  is of the form
!
1 0
:
2 1
Thus, we have shown that the monodromy representation is uniquely determined up to con-
jugation in SL2 (Z). Notice also that according to Equation 6 the magnetic charge nm of the
hyper-multiplet that is becoming masses at 2 is 1. The same of course holds at 2 . In fact,
because the SL2 (Z) is uniquely determined up to conjugation it is invariant, up to conjugation,
under the symmetry u 7! u.
It is easy to identify the family of elliptic curves over the twice punctured plane with this
monodromy. The base of the family can be identied with the modular curve H= (2) given by
dividing out the upper half-plane by the (free) action of the subgroup of PSL2 (Z) consisting of
all matrices congruent to the identity modulo two. The family is the universal family of elliptic
curves with `level two structure' over this base. Since the singularities are at u = 2 , the
Weierstrass equation for the family of elliptic curves associated to this quantum theory is

y2 = (x u)(x 2 )(x + 2 ):

Now let us consider the two-form  as described in the last lecture. We know that in the
Weierstrass model given above  has the general form

 = f(u)du dx
y:
Implicitly in the discussions before we assumed that  was a complex symplectic form without
zeros or poles at least away from the exceptional values of u. Such zeros or poles would produce

15
non-physical singularities. If we take f to be a non-zero constant, then the Kahler form  ( ^ )
is smooth except at 1; 2 where it has the correct singular behavior. Any other f would
produce either poles or zeros at other points or the wrong singular behavior at one of 1; 2.
It follows that  = Cdu dxy for an appropriate constant which can be computed by comparing to
the formula a2 = u near u = 1.
Clearly,    
udx
 = d C y = d C x2 4 : 2ydx

The form  = 2Cydx=(x2 4) is a dierential of the second kind on the total space E of the
family of elliptic curves. This means, as we have been asserting all along, that [] 2 H 2 (E ; C) is
trivial and hence that the monodromy of our family of representations of the theory is contained
in SL2 (Z), inside the bigger group ISL2 (Z). This was expected since the theory has no conserved
charges except ne ; nm .

I.7 Description of the BPS spectrum

The line in the u-plane where a=aD is real is a simple closed curve that passes through the points
u = 2 . The spectrum of BPS states is continuous o this circle. Approaching from u = 1,
we see that at u = 2 we have a = 0 and aD 6= 0, whereas at u = 2 we have a + aD = 0.
(Of course, the exact values depend on the path chosen from 1 to 2, but if we come in along
rays from innity toward the origin then given equations hold.) Since we can approach the two
singularities from innity without crossing the jumping locus for the BPS states, we see that the
BPS states that are becoming massless at u = 2 are indeed among the original BPS states
at innity. That is to say, these particles are analytic continuations of the original magnetic
monopoles at innity.
Now let us examine whether, at other points on the jumping locus besides 2 , there are
particles forced to become massless. As we move along the jumping line on the open upper arc
from 2 to 2, we nd that for every s; 0 < s < 1 there is a point where sa aD = 0. So if
there is a BPS state at innity has charges (ne ; nm ) with (ne ; nm) = t(s; 1) for some t 2 R,
then at some point along this open arc we nd that this state becomes massless. This, and the
symmetric argument for the open lower arc of the jumping line, show that all the BPS states at

16
innity that satisfy
(ne ; nm ) = (p; q)
with 0 < p < q become massless somewhere on the jumping curve minus 2 . But we know by
our analysis of the singularities that there are no points on these open arcs where BPS states
(or any other charged states) become massless. The conclusion is that there are no BPS states
at innity satisfying the above inequality. Said another way the only BPS states at innity are
those of charges
(ne; nm ) = (1; 0)
(ne ; nm) = (p; 1):
Of course, these are exactly the BPS states that do exist at innity. The rst are the vector
multiplets and the second are the magnetic monopoles. This gives us a check of the internal
consistency of the theory.
This analysis of the BPS spectrum in the quantum theory can be compared in an interesting
way to the following classical computation. Let Mk be the moduli space of k-monopole solutions
of this theory, reduced by dividing by translations. It is a hyperkahler manifold of dimension
4k 4 that has been much studied mathematically. It can be shown that L2 -holomorphic forms
(possibly valued in a at line bundle) on Mk would give k-monopole bound states, for k > 1.
From what we have seen, there should be no such bound states, so there should be no L2 -
holomorphic forms on Mk . More generally, an argument by Sen using duality of N = 4 super
Yang-Mills (rather than the N = 2 theory that we have considered in this lecture) determines
the full L2-cohomology of the Mk .

17